Interference Cancellation for Downlink MU-MIMO November 2009 doc.: IEEE 802.11-09/1234r0doc.: IEEE 802.11-yy/xxxxr0 doc.: IEEE 802.11-yy/xxxxr0 March 2010 Interference Cancellation for Downlink MU-MIMO Date: 2010-03-15 Authors: Sameer Vermani, Qualcomm Sameer Vermani, Qualcomm

doc.: IEEE 802.11-09/1234r0doc.: IEEE 802.11-yy/xxxxr0 November 2009 doc.: IEEE 802.11-09/1234r0doc.: IEEE 802.11-yy/xxxxr0 doc.: IEEE 802.11-yy/xxxxr0 March 2010 Abstract Downlink (DL) Multi-user (MU) MIMO is identified as a key technology to improve the overall network performance In 09/1234r0 we showed that : Interference Cancellation (IC) at the STA makes downlink (DL) MU-MIMO more robust To support Interference Cancellation in DL MU-MIMO at the STA: AP must transmit enough LTFs to enable channel estimation for the total number of spatial streams in the DL We call this mode of LTF transmission the ‘Resolvable LTF’ mode AP must signal to each STA which spatial streams are meant for it This document is a review of IC concept in 09/1234r0 with an additional strawpoll at the end Sameer Vermani, Qualcomm Sameer Vermani, Qualcomm

March 2010 Outline Introduction Interference Cancellation Receive processing Sources of CSI Error at AP Simulation results for 40MHz and reasonable product configurations AP with 4Tx; Clients have 2 Rx AP with 8Tx; Clients have 3 Rx Conclusions Straw poll Sameer Vermani, Qualcomm

Introduction to Interference Cancellation March 2010 Introduction to Interference Cancellation In DL MU-MIMO, clients can have more receive (Rx) antennas than the number of spatial streams they receive The additional antennas can be used for Interference Cancellation (IC) / Interference Suppression Particularly useful when precoding is imperfect due to errors in the CSI available at the AP This calls for a DL MU-MIMO preamble design that can support IC Each client should receive as many LTFs as needed to train the total number of spatial streams in the DL Each client should know which spatial streams are meant for it Sameer Vermani, Qualcomm

Receive MMSE for Interference Suppression March 2010 Receive MMSE for Interference Suppression For instance, consider a 4-antenna AP transmitting 1 ss each to 4 STAs each with 2 Rx antennas, the Rx signal at 8 Rx antennas is given by: The equivalent precoded channel is Hequiv = H8x4W4x4 The first two rows of Hequiv is the channel seen by STA1; H1 = Hequiv(1:2,:) STA1 can do the following MMSE processing to reduce the interference from other STAs: where the first element of x1 gives the estimate of the symbol for STA1 and 12 is the noise variance at STA1 Sameer Vermani, Qualcomm

Sources of CSI Errors at AP March 2010 Sources of CSI Errors at AP Pathloss to the STA or the amount of quantization in the CSI feedback report The channel estimation SNR or quantization level is fundamental to the accuracy of CSI Time variations in the channel A non-zero time interval between CSI feedback and DL MU-MIMO transmission causes discrepancies between precoding weights and the actual channel Feedback delay of 20 ms results in an error floor of -25 dBc (assuming a coherence time of 800 ms) Modeled as two independent additive noise sources in the CSI CSI Feedback Delay Error Floor {-20, -25, -30} dBc Channel Estimation Error Floor (Pathloss dependent) At high SNRs, CSI feedback errors dominate and at low SNRs, pathloss errors dominate Sameer Vermani, Qualcomm

March 2010 Simulations Determine the gains of using MU-MIMO and Interference Cancellation (IC) We plot the 10 percentile and 50 percentile points from the CDF of the aggregate PHY throughput (measured at the AP) as a function of pathloss For comparison, we also plot the corresponding sequential beamforming (BF) data quantities SVD based transmission with equal MCS per spatial stream Data rates averaged across sequential transmissions to the clients Sameer Vermani, Qualcomm

Results for 4 antenna AP, Four clients each with 2 Rx, full loading March 2010 Results for 4 antenna AP, Four clients each with 2 Rx, full loading Sameer Vermani, Qualcomm

Simulation Parameters March 2010 Simulation Parameters AP with 4 Tx antennas transmitting at 24 dBm Noise floor of -89.9 dBm 4 STA with 2 Rx antenna each -35 dBc of TX distortion Equal Pathloss to each STA, varied from 70 to 95 dB Single SS per STA in the MU-MIMO case and 2 ss for Tx BF case TGac Channel Model D, NLOS Results for 200 channel realizations For MU-MIMO, MMSE precoding done to beamform the 1 ss of each STA to one of its antennas Two sources of CSI error at AP Channel estimation floor at client = -(Total Tx Power – Pathloss + 89.9 dBm (Thermal noise)) Feedback delay error = {-20, -25 ,-30} dBc Sameer Vermani, Qualcomm

4 antenna AP, Four 2 Rx clients, -20 dBc feedback error March 2010 4 antenna AP, Four 2 Rx clients, -20 dBc feedback error Eigen BF TDMA MU-MIMO w/o IC MU-MIMO with IC Eigen BF TDMA MU-MIMO w/o IC MU-MIMO with IC MU-MIMO with IC gives best performance Interference Cancellation improves performance for a poor CSI accuracy IC enables full loading Compare with slide 22 in Appendix, which shows the 3 ss results Performance better with 3 ss in the absence of IC Sameer Vermani, Qualcomm

4 antenna AP, Four 2 Rx clients, -25 dBc feedback error March 2010 4 antenna AP, Four 2 Rx clients, -25 dBc feedback error Eigen BF TDMA MU-MIMO w/o IC MU-MIMO with IC Eigen BF TDMA MU-MIMO w/o IC MU-MIMO with IC For all pathlosses between 70 and 95, MU-MIMO with IC gives substantial gains Sameer Vermani, Qualcomm

4 antenna AP, Four 2 Rx clients, -30 dBc feedback error March 2010 4 antenna AP, Four 2 Rx clients, -30 dBc feedback error Eigen BF TDMA MU-MIMO w/o IC MU-MIMO with IC Eigen BF TDMA MU-MIMO w/o IC MU-MIMO with IC For all pathlosses between 70 and 95, MU-MIMO with IC gives best performance Gains of IC reduce as CSI accuracy improves Sameer Vermani, Qualcomm

Results for 8 antenna AP, Three clients each with 3 Rx March 2010 Results for 8 antenna AP, Three clients each with 3 Rx Sameer Vermani, Qualcomm

Simulation Parameters March 2010 Simulation Parameters AP with 8 Tx antennas transmitting at 24 dBm Noise floor of -89.9 dBm 3 STA with 3 Rx antenna each -35 dBc of TX distortion Equal Pathloss to each STA, varied from 70 to 95 dB Two SS per STA in the MU-MIMO case and 3 ss for Tx BF case TGac Channel Model D, NLOS Results for 200 channel realizations For MU-MIMO, MMSE precoding done to beamform the 2 ss of each STA to two of its antennas Two sources of CSI error at AP Channel estimation floor at client = -(Total Tx Power – Pathloss + 89.9 dBm (Thermal noise)) Feedback delay error = {-20, -25 ,-30} dBc Sameer Vermani, Qualcomm

8 antenna AP, Three 3 Rx clients, -20 dBc feedback error March 2010 8 antenna AP, Three 3 Rx clients, -20 dBc feedback error MU-MIMO with IC gives best performance IC improves performance for a poor CSI accuracy Sameer Vermani, Qualcomm

8 antenna AP, Three 3 Rx clients, -25 dBc feedback error March 2010 8 antenna AP, Three 3 Rx clients, -25 dBc feedback error For all pathlosses between 70 and 95, MU-MIMO with IC gives best performance Sameer Vermani, Qualcomm

8 antenna AP, Three 3 Rx clients, -30 dBc feedback error March 2010 8 antenna AP, Three 3 Rx clients, -30 dBc feedback error Gains of IC reduce here Precoding is very good Sameer Vermani, Qualcomm

Conclusions IC makes MU-MIMO robust to poor CSI accuracy at the AP March 2010 Conclusions IC makes MU-MIMO robust to poor CSI accuracy at the AP Significantly improves PHY throughput Enables fully loaded MU-MIMO This calls for a DL MU-MIMO preamble design that can support IC AP must transmit enough LTFs to enable an STA to train the total number of spatial streams in the DL AP must signal to each STA which spatial streams are meant for it Sameer Vermani, Qualcomm

March 2010 Straw Poll Do you support the Interference Cancellation concept described in this document by inclusion of the following section and text in the Tgac spec framework document: “4.1 Resolvable LTFs for DL MU-MIMO In a DL MU-MIMO transmission, LTFs are considered “resolvable” when the AP transmits enough LTFs for an STA to estimate the channel to all spatial streams of every recipient STA. In order to enable interference cancellation at an STA during a DL MU-MIMO transmission, an AP may transmit the preamble using resolvable LTFs. ” Yes No Abstain Sameer Vermani, Qualcomm

March 2010 Appendix Sameer Vermani, Qualcomm

Methodology used to get to Data Rate CDFs March 2010 Methodology used to get to Data Rate CDFs For each spatial stream Calculate the post processing SINR on each tone Map the post processing SINR to capacity using log(1+SINR) Average the capacity across tones to get Cav Use Cav to calculate SINReff using Cav = log(1+ SINReff) Map the SINReff to a rate using the AWGN rate table This method is used in other WAN standards, e.g., 3GPP2 Sum the rate across all spatial streams for one channel realization to get to aggregate PHY throughput Do this for 200 channels to get to the CDF of aggregate PHY throughput Sameer Vermani, Qualcomm

4 antenna AP, Three 1x1 clients, -20 dB feedback error March 2010 4 antenna AP, Three 1x1 clients, -20 dB feedback error 70 75 80 85 90 95 100 200 300 400 500 600 700 800 Pathloss in dB PHY Rate in Mbps measured at AP Variation of 10 percentile PHY Rates with pathloss Eigen BF TDMA MU-MIMO w/o IC MU-MIMO with IC 70 75 80 85 90 95 100 200 300 400 500 600 700 800 Pathloss in dB PHY Rate in Mbps measured at AP Variation of 50 percentile PHY Rates with pathloss Eigen BF TDMA MU-MIMO w/o IC MU-MIMO with IC For all pathlosses between 70 and 95, MU-MIMO gives substantial gains IC curve lies on top of MU-MIMO w/o IC In absence of IC, 4 SS MU-MIMO performs worse than 3 SS MU-MIMO Compare green curve of this slide with blue curve of slide 10 Better to transmit at 75% loading in the absence of extra antenna at the STAs Scheduler decision Sameer Vermani, Qualcomm